Current large detector-array technologies, for particle or photon imaging or for counting, are frame rate or count rate limited. Examples of these detector array technologies are focal plane arrays used for visible and infrared light imaging and detector arrays used for x-ray florescent spectroscopy. The imaging speed limitation is usually related to the need to digitize all the data in the array and transfer it to a digital computer for processing. Digitizing and transferring all the data produces a data rate bottleneck and digitally processing all the data produces a processing rate bottleneck. High frame rates or count rates can currently be handled by massively-parallel digital processors. However, because of the shortcomings of these systems, namely high cost, large size and large-power consumption, this solution is not in common use.
This invention relates to cameras or sensors using detector arrays and analog-processing integrated-circuit architectures; the analog-processing circuitry allows a very compact and low power solution to the high-frame-rate and counting-rate problem. Data from the detector arrays is output, in analog form, directly to analog processing array chips where a processing algorithm or algorithms are implemented in unit cell circuitry. At the highest speeds, there is one or more processing unit cells for each detector and each analog-processing chip is composed of an array of unit cells. Processing occurs simultaneously, that is in parallel, in all unit cells. The analog processing chips provide for offset and gain correction; in the conventional imager design this correction is made in the digital computer. The analog-processing unit-cell-array chips and detector array are designed as a single camera or sensor system. Only data that satisfies the algorithms is output from the sensor greatly reducing the data rate. Special circuitry is used to select only the data that has satisfied the algorithm. The camera or sensor is called a smart camera or sensor because it processes data and selects the data to be output. Because most or all of the data processing occurs in the unit-cell arrays and because of the reduced data rate, the sensor output data can be further processed, if necessary, by a conventional digital computer without data-processing or data-rate bottlenecks. Since the camera or sensor is only composed of a few chips it is very compact and consumes little power.
There is currently a very complex analog processing architecture, using mammalian retina circuit analogs that propose to solve the high-data-rate, image-processing problem. The architecture, incorporating z-plane technology, uses thinned chips, usually silicon (one chip for each column), butted up on edge against the backside of the detector array chip. One edge of each z-plane chip is connected to each detector in a single column in the detector array. The detector array then sits on top of a cube of semiconductor chips. The processing algorithm is that of Carver Mead""s biological-analogy Silicon Retina1,2. The shortcomings of z-plane technology are: (1) it requires a large volume of silicon and hence is expensive (512 chips for a 512xc3x97512 detector array for example); (2) the z-plane chips must thinned by a special process and all interconnected; this is expensive and difficult to accomplish without error; (3) image resolution is limited by the thickness of the z-plane chips; (4) difficulties dissipating heat in the z-plane three-dimensional cube limits processing speed; and (5) the number of outputs are very large, equal or greater than the number of detectors. Limitations of the biological-analogy algorithm are: (1) spatial averaging reduces resolution to larger than the pixel size; (2) difficulties with implementing gain and offset corrections leads to spurious images and noise; and (3) the silicon-retina edge algorithm is of limited application.
1 Carver Mead, Analog VLSI and Neural Systems, Addison Wesley 1989. 
2 M. A. Mahowald and C. Mead, The Silicon Retina, Scientific American, May 1991, pg 76. 
The current invention overcomes all the difficulties of the z-plane technology because it requires very few chips and uses conventional CMOS without special processing. There are no limitations in detector size; the two dimensional architecture easily dissipates heat. There are few outputs and resolution is defined by the pixel geometry. Gain and offset corrections are straight forward and the algorithms that can be implemented are quite general.
Another application of the present invention is a High-Count-Rate Energy-Discriminating x-ray photon or high energy particle Detector (HRED) which can process three to four orders-of-magnitude higher count rate than current energy-dispersive spectrometers. High-count-rate x-ray detectors are important in many areas of science where there is a large background and high signal to noise ratio is obtained by energy discriminating the signal from the background. Examples are the non-invasive high-speed quantitative measure of lead in bones and other elements in other organs3,4,5. Data collection is particularly limited in structural biology investigations of dilute samples where detectors have not kept pace with synchrotron source development. The Extended X-ray Absorption Fine Structure (EXAFS) technique, counting fluorescent x-rays, has been known for some time6, for example, but the counters have very limited data rates which cannot adequately take advantage of current and future synchrotron source fluxes. Many samples have low concentrations of the element of interest that is embedded in a matrix of energy absorbing molecules. Under these conditions conventional detectors expend their count rate separating the desired-element-fluorescence x-rays from the larger number of quasi-elastic and matrix-fluorescence x-rays.
3. L. Preiss and M. A. Tariq On the use of L x-ray fluorescence for bone lead evaluation, Radiocanal. Nucl. Chem. Let. 164 (6), 381-385 (1992). 
4. I. L. Preiss and T. PTAK, Trace Element Profiles of Biological Samples Using Radioscope X-ray Fluorescence, Nuclear Instruments and Methods in Physics Research A242 (1986) 539-543. 
5 A. S. Frank et al, Trace Element Profiles in Murine Lewis Lung Carcinoma by Radioscope-Induced X-Ray Fluorescence, AJP, March 1986, Vol. 122, No. 3. 
A widely used x-ray fluorescence-detected spectroscopy detector is the 13-element Canberra Ge Detector7. Reference 7 shows the count rate of a single element, using a 1 xcexcs shaping time corrected for dead time losses, is about 2xc3x97105 photons/sec (dead time of about 5 xcexcs). This count rate is inadequate for many synchrotron-based experiments and is a source of frustration for researchers who are preparing to use the newest beam lines. Furthermore the shaping time for optimum energy resolution8 is about 14 xcexcs and leads to about an order of magnitude reduction in counting rate at the optimum energy resolution. A single module of the current invention is roughly the same area as a single element of the 13-element detector and has a count rate of 2.5xc3x97108 secxe2x88x921 at the highest energy resolution. This count rate is three to four orders of magnitude higher than a single element of the 13-element detector.
6 I. J. Jaklevic et al, Solid State Communications, 23, 679 (1977). 
7 S P. Cramer et al, A 13-Element Ge Detector For Fluorescence EXAFS, Nuclear Instruments and Methods in Physics Research, A266, 586 (1988). 
8 Quantitative X-ray Spectrometry, Ron Jenkins et. al., Marcel Dekker, Inc., 1981. 
The present invention comprises a sensor for detection of photons or particles. Detection takes the form of a two-dimensional unit-cell array of electrical signals. The sensor includes an analog image processor for implementing an image-processing or energy-discrimination algorithm and for outputting only the signals and/or the positions of the unit cells for which the algorithm was satisfied; an alternative output is the number of unit cells which satisfy the algorithm. One embodiment of the invention also comprises an analog correction processor, located between the sensor and image processor, that corrects the signals for nonuniformities of the invention. The invention further comprises drive electronics to provide the timing pulses and biases necessary for operation of the sensor, output electronics for conversion of the analog to digital signals and storage of the digital signals, and digital data processors to further analyze the digital signals.
It is the object of the present invention to provide high-speed devices for the collection and analysis of photon or particle, image or counting data which overcomes problems of prior systems associated with data bottlenecks, slow image processing and slow counting.
It is further the object of the present invention to provide a device for making high-speed analog gain and offset corrections to analog data.
It is still further the object of the present invention to provide a device for high-speed implementation of data processing algorithms on analog data and for outputting only the important data.